A major problem when evaluating the accuracy of quantitative computed tomography (QCT) against chemical analysis of bone mineral is to ensure that the same region of interest (ROI) is measured with two techniques. The method described here attempts to reach this objective. Forty-nine lumbar vertebrae were scanned. The mid-vertebral portion was sawed and embedded in a polyester resin. The ROI defined with an automated contour detection was transmitted, using a Hewlett-Packard scanjet, to a computer-assisted milling machine which excised a trabecular ROI sample. The volume of each trabecular sample was determined using an Archimedian method. ROI volumes assessed on QCT were highly correlated with those obtained with the Archimedian method (r = 0.98). The standard error of the estimate was 170 mm3 (4.5%). These results suggest that trabecular samples obtained with this method optimally agree with the ROIs selected on QCT examination.